Artificial intelligence is increasingly influencing how breast cancer screening is approached, with growing attention on risk stratification alongside image interpretation. Advances in data-driven analysis are shifting the focus from population-based screening intervals towards more individualised assessments that may inform supplemental imaging and follow-up strategies. At a recent scientific meeting of the European Society of Breast Imaging, held in cooperation with the British Society of Breast Radiology, clinicians and researchers discussed how AI models built on mammography data could help identify women at higher short-term risk of breast cancer. The discussions highlighted both the promise and the limitations of current approaches, addressing how AI compares with established risk models and where further clinical evidence is required. These perspectives reflect an evolving effort to use existing screening infrastructure not only for detection but also for more precise risk prediction and treatment planning.
Limitations of Traditional Risk Models
Established breast cancer risk models have been used for many years and are considered effective in assigning women to broad risk categories. Their calibration allows for identification of general risk classes, but several practical and clinical limitations remain. Uptake across screening populations has been relatively low, and many models rely heavily on family history rather than broader indicators of lifetime or near-term risk. This emphasis can reduce their usefulness for women who develop interval cancers, which arise between scheduled screening examinations.
Must Read: Mammography-Linked Lung Screening Increases Uptake
Additional challenges relate to the collection and completeness of data required for these models. Gathering detailed personal and family history information can be difficult in routine clinical settings, leading to assessments based on incomplete inputs. Performance has also been shown to vary across ethnic subgroups, introducing concerns about bias and unequal applicability. These factors limit the suitability of traditional approaches for guiding short-term clinical decisions, particularly when the aim is to identify women who may benefit from changes in screening frequency or modality.
AI-Based Risk Stratification Using Mammography
Research presented at the meeting described efforts to expand risk assessment beyond primary prevention by using mammography data already generated within screening programmes. The objective is to assess risk within a narrower clinical timeframe, enabling identification of women who could benefit from supplemental screening or shorter screening intervals, while also reducing false positive predictions. Rather than relying solely on background risk factors, AI models analyse imaging features to support risk prediction and classification.
An example was presented in which women with moderate risk identified at one time point were later flagged by AI tools as having higher risk, several years before a subsequent diagnosis. Such an approach aims to highlight clinically actionable cases earlier, potentially improving outcomes by facilitating timely intervention. A key feature of these AI models is their ability to identify high-risk individuals regardless of mammographic density, a factor that has traditionally complicated risk assessment and detection.
Interval cancers account for a substantial proportion of breast cancer diagnoses, and the use of AI within a narrow time window may increase the likelihood of detecting these cancers, although some increase in false positive recalls may occur. Validation efforts across multiple screening settings have shown promising performance, with AI models consistently outperforming traditional risk models. Ongoing work focuses on identifying and addressing confounding factors to further improve accuracy and robustness.
Response Prediction and Clinical Integration Challenges
Beyond risk stratification, AI is also being explored for its potential role in predicting response to treatment and supporting diagnostic decisions. Image-based AI has been shown to improve prediction of pathologic complete response compared with models based solely on clinical features, although the magnitude of improvement remains modest. Interest has also emerged around the possibility of using AI to support selection of less toxic therapeutic regimens for certain patients.
In post-treatment settings, AI may help identify individuals who do not require surgery or radiotherapy, as well as those at higher risk of poorer prognosis who may benefit from additional intervention. Despite these possibilities, research in this area remains limited, and much of the work is still in early stages. The absence of established clinical guidelines continues to be a major barrier to broader implementation. Clinical trials are underway to determine how many women would be classified as high risk using AI models and to assess the impact on rates of interval and advanced cancers. Consideration is also being given to the psychological effects of risk assessment and its influence on screening attendance.
The integration of artificial intelligence into breast cancer screening is extending the role of mammography from detection towards more personalised risk prediction and, potentially, treatment planning. AI-based models offer advantages over traditional risk assessments by using existing imaging infrastructure and performing consistently across different breast densities. At the same time, applications such as response prediction remain under development, with modest gains and limited clinical evidence to date. The discussions highlighted the need for robust clinical trials and clear guidelines before these tools can be adopted more widely. The emerging evidence underscores both the potential of AI to support more precise screening strategies and the importance of careful evaluation as these technologies move closer to routine clinical use.
Source: Healthcare in Europe
Image Credit: iStock
